Page 11 - November Merged
P. 11
Feature | Stop using ChatGPT
BEST CHATBOT FOR ANALYSING SPREADSHEET DATA
To test how well our AI assistants can graphs showing the average World Cup
analyse data, pick out the key points and attendance and number of goals over
judge their significance, we uploaded a time. It highlighted the highest-scoring
spreadsheet of historical FIFA World Cup matches and teams, and attributed the
data. This publicly available dataset decrease in goal rates to the evolution
contained details of all World Cup matches of tactical sophistication and physical
from 1930 to 2022, including the teams, fitness. But its response was brief and not
rounds, scores (full-time and half-time), particularly incisive.
stadiums, attendance and more. Perplexity identified the same key
trends as the other AI tools and delivered
PROMPT its summary in a dry, methodical tone.
Analyse this spreadsheet of every World For example: “the attendance trend is one
Cup match from 1930 onwards and of the clearest structural changes”. It
identify the key trends. Claude generated an infographic and a text offered to create a table organising the
summary of the uploaded data. data by tournament year, stage or team,
ChatGPT provided several insights into but then said it couldn’t because we’d hit
the World Cup data we uploaded, in its concise but interesting, such as 7.6 the file-analysis limit.
customary bullet-pointed style. It percent of matches going to extra time or
identified that matches have become penalties. It also mentioned the ‘Mineirazo’ Winner: Claude
lower-scoring; draws and close results semi-final of the 2014 tournament, when
have become more common; and Germany thrashed Brazil 7-1. Claude dug the deepest
home-side advantage is “surprisingly Copilot’s overview of our spreadsheet into our spreadsheet, and
strong”. It concluded that World Cups seemed rather obvious – more matches, presented its analysis of the
have become far more defensive and larger attendance, fewer wild scorelines World Cup data in the most engaging
tactically balanced. – but its actual analysis surpassed manner. Copilot looked beyond the
Claude exceeded our expectations by ChatGPT’s. It noted the “tug-of-war for spreadsheet to get extra information
using a graph, bar chart and table to dominance between Europe and South from other sources, whereas ChatGPT
display the World Cup data. This included America”, the rise of extra time, and the and Gemini just used what we’d
the total number of matches and goals increase in penalty shootouts after they provided, and Perplexity sounded
(see screenshot) and the home-win rate. were introduced in 1982. bored by the prompt.
Its text summary of key trends was Similar to Claude, Gemini created
BEST CHATBOT FOR RECOMMENDATIONS
PROMPT Copilot recommended the GMKtec article titled ‘Best mini PC under $1,000’ on
Recommend a mini PC that costs less NucBox M2 Pro S – a newer and more Geekom’s own website. The Beelink model
than $1,000, has at least 16GB of RAM, powerful mini PC than ChatGPT’s is now three years old and we couldn’t find
1TB of storage and a speedy processor, suggestion, run by an Intel Core i7 chip. It any store still selling it in Australia.
and includes plenty of ports. briefly explained why this model stands Despite claiming to have used 10 sources,
out, and suggested two alternatives, Perplexity’s response was based on the
ChatGPT gave our query proper mislabelling one ‘Fast work mini PC’. same article as Gemini – so it’s not
consideration, outlining the main things Frustratingly, Copilot didn’t provide links surprising it recommended the Geekom
to look for in a mini PC for under $1,000. to manufacturers’ websites or Amazon – A6. It didn’t suggest any alternatives and
It recommended the GMKtec NucBox M5 just to a store we hadn’t heard of. its response was just a short list of specs.
Ultra as the ‘Best overall’ – and provided a Gemini confidently recommended two
direct link to buy it on Amazon (see “excellent” mini PCs that met our Winner: ChatGPT
screenshot). Helpfully, ChatGPT requirements: the Geekom A6 and the
suggested alternatives for best value, Beelink SER5 Max 5800H. The former is There was no contest in this
upgradeability and ‘premium-feeling’, indeed a great choice, but Gemini’s round: ChatGPT gave us the
accompanied by photos and a table reasoning drew solely – and lazily – on an most reliable and well-reasoned
comparing the specs of all four models. recommendations. It offered
Claude confused us by assuming we suggestions for different mini-PC
were asking specifically about the Beelink needs, detailed their specs and
SER8 mini PC. Its sole source of info was provided links to buy them – unlike
Amazon and its answer was surprisingly Copilot. Claude seemed out of its
poor. It described the SER8 as being comfort zone, while Gemini and
“widely available on Amazon” but when Perplexity were too lazy to properly
we clicked through to Amazon the mini ChatGPT recommended four mini PCs and research our query.
PC’s listing said ‘Currently unavailable’. provided links to buy them.
OCTOBER 2026 APC MAGAZINE 41

